emiliofidalgo / ibow-lcd Goto Github PK

Appearance-based Loop Closure Detection using Incremental Bags of Binary Words

License: GNU General Public License v3.0

CMake 4.86% C++ 60.61% MATLAB 34.53%

ibow-lcd's Introduction

iBoW-LCD: Appearance-based Loop Closure Detection using Incremental Bags of Binary Words

iBoW-LCD is a library that can be used to detect loop closures in a sequence of images. Instead of using the typical Bag-of-Words (BoW) scheme, iBoW-LCD makes use of an incremental Bags of Binary Words algorithm called OBIndex2, allowing the use of binary descriptors to accelerate the image description process and avoiding the problems that classical BoW approaches present.

iBoW-LCD is released as a ROS package, and relies on OpenCV 3.x and Boost libraries as well as OBIndex2 package. It can be used with any binary descriptor computed using the OpenCV format.

Note that iBoW-LCD is research code. The authors are not responsible for any errors it may contain. Use it at your own risk!

Conditions of use

iBoW-LCD is distributed under the terms of the GPL3 License.

Related publication

The details of the algorithm are explained in the following publication:

iBoW-LCD: An Appearance-based Loop Closure Detection Approach using Incremental Bags of Binary Words
Emilio Garcia-Fidalgo and Alberto Ortiz
IEEE Robotics and Automation Letters, Vol. 3, No. 4, Pgs. 3051-3057 (Oct. 2018)

A preprint can be found here. If you use this code, please cite:

@article{Garcia-Fidalgo2018, 
  author={Emilio Garcia-Fidalgo and Alberto Ortiz}, 
  journal={IEEE Robotics and Automation Letters}, 
  title={iBoW-LCD: An Appearance-Based Loop-Closure Detection Approach Using Incremental Bags of Binary Words}, 
  year={2018}, 
  volume={3}, 
  number={4}, 
  pages={3051-3057}, 
  doi={10.1109/LRA.2018.2849609}, 
  month={Oct}
}

Installation

First of all, you have to install OBIndex2 and its dependencies as a ROS package.
Clone the repository into your workspace:

cd ~/your_workspace/src
git clone http://github.com/emiliofidalgo/ibow-lcd.git

Compile the package using, as usual, the catkin_make command:

cd ..
catkin_make -DCMAKE_BUILD_TYPE=Release

Finally, you can run an example with:

rosrun ibow-lcd demo /directory/of/images

Usage

To see an example of how to use the loop closure detector, see the demo file src/main.cc.

Contact

If you have problems or questions using this code, please contact the author ([email protected]). Feature requests and contributions are totally welcome.

ibow-lcd's People

Contributors

Stargazers

Watchers

ibow-lcd's Issues

Code is too slow?

Hi,
I came across your IROS paper and found it interesting. I could compile your code and run it on my images.

I modified your main.cc file a bit:
I am extracting 500 ORB features instead of your 1500.
My image size : 640x512.

I am timing the code and I find that this part, takes up more than 1.5 sec.

lcdet.process(i, kps, dscs, &result);

I am wondering is this normal or I am making a mistake running your code?
I did a catkin_make with Release flag as you have mentioned in this readme on this repo.

Precision-recall different curve on different runs

Hi @emiliofidalgo,

I am able to run the code successfully on the Lip6Indoor dataset. But every time I evaluate (using rosrun ibow-lcd evaluator $Path$/evaluation/configs/debug/config_Lip6In.json), I get different data in loops.txt file.

Is it because of randomness in feature extraction? Otherwise what causes this variance?

This leads to a different PR curve (inferior recall performance) as compared to the one given in the paper.

This is what I get in one of the run:

While in the paper it is something like this:

Is it supposed to be varying that much or I am doing wrong?

Couldn't find executable named demo

Hi, I am new to this and came across ibow-lcd during a project and want to try it out but I've ran into a problem.
I did exactly as the installation guide but when i try to run the demo by using [rosrun ibow-lcd demo /directory/of/images],
ROS returns [[rosrun] couldn't find executable named demo below /home/robo/catkin_ws/src/ibow-lcd]
Any idea on what might be wrong?

I'm using ros kinetic on ubuntu 16.04.6.